Pro Skill Packs logo PRO SKILL PACKS

Guides / Marketers and agencies

How to write ad copy variants you can actually test

Build a grid of ad variants that differ in one thing each, so a test tells you something. Includes a prompt that stays inside your real offer and the character limits.

Ten ads that all say roughly the same thing teach you nothing. If one wins, you cannot tell why. A useful test changes one thing at a time and is built from facts about your offer that are true.

Step 1: write down the facts

List what you can honestly say: the product or service, who it is for, the price or offer, what is included, any real proof you can show (a rating you can link, a number of years, a guarantee in your terms), and what you will not claim. Anything not on the list cannot appear in an ad.

Check the rules for your ad platform on claims and restricted content before you launch. They differ and they change.

Step 2: pick the angles

An angle is the reason to click. Choose three or four that your facts support, for example:

  • Problem. Names what is annoying them now.
  • Result. Names what they get.
  • Proof. Names a real, checkable detail.
  • Offer. Names the price, deal or guarantee.
  • Objection. Answers the doubt, such as "no contract".

Step 3: use the prompt

Write ad copy variants for a test. Use ONLY the facts below. Do not add claims, numbers, awards, testimonials, discounts or guarantees that are not listed.

Business and offer: [one or two sentences]
Audience: [who]
Facts I can back up: [bullet list]
Things I must not say: [bullet list, for example no medical claims, no "best"]
Platform and limits: [for example Google responsive search ad: headlines up to 30 characters, descriptions up to 90]

Angles to cover: [for example problem, result, proof, offer]

For each angle write 3 variants. Change only the wording of that angle, keep the call to action the same across the grid. Present it as a table with columns: angle, text, character count, which fact it uses.

Check each variant against the character limit and count the characters yourself. Mark any that are over. Do not use exclamation marks, all capitals, or words like "best", "guaranteed" or "#1" unless I listed that exact claim as a fact.

At the end, list any variant that relies on something I did not give you.

Step 4: check before you run

  • Count the characters yourself for any that look close. Assistants miscount.
  • Match each variant to a fact on your list. If you cannot, delete it.
  • Make sure the landing page says the same thing as the ad.
  • Remove variants that are the same idea in different words. You want real differences.

Step 5: run a test you can read

Keep one thing different between the ads you compare. Give them the same audience, budget and time. Decide in advance what counts as a win, such as cost per sign-up, and how many results you need before you judge. A handful of clicks proves nothing.

When one angle wins, write the next round around that angle and change something else, such as the opening words.

Limits

  • The assistant has no data about your audience. The grid is a set of hypotheses, not predictions.
  • It cannot know your platform's current policies. Read them.
  • Small budgets give noisy results. Do not read too much into small differences.
  • It writes text only. Images, targeting and the landing page matter at least as much.
  • Never use a testimonial, rating or number you cannot show on request.

Browse all skills in the catalog